Keynote lecture

Formulation design has not kept pace with advances in drug discovery, and most development programs still rely on empirical search over narrow design spaces, which too often yields “the best of what we tried” rather than what is truly optimal. The formulation of therapeutic agents can significantly improve their safety and efficacy. However, the design and development of drug formulations remains expensive, labour-intensive and time consuming, with a heavy reliance on the expertise of the formulation development team and on compositions of formulations that have been approved to date. In the design of these systems, there are numerous parameters that must be considered in relation to the drug, material(s) or excipient(s), as well as processing variables. Experimental evaluation of every combination is intractable and at this time it is not possible to predict the performance of specific formulations a priori. As a result, it is likely that some of the formulation candidates that have moved forward to clinical development are not optimal but rather the best that could be achieved with the time and resources available. Here, we present AI-enabled workflows that reframe formulation as a rigorous optimization problem and shift the objective toward “the best of what is possible.” Machine learning (ML) has led to significant advances in fields such as drug discovery and materials science. We integrate multiple ML approaches to map relationships between composition, properties and performance and to directly guide the selection of formulations that match a patient-centred target product profile in minimum time. To support this, we have implemented a self-driving laboratory (SDL), in which AI proposes experiments and roboticssupported automation executes them, enabling rapid, reproducible collection of robust data to train and refine our models. Our systems are designed from the outset to capture end-to-end metadata, including raw material attributes, process parameters, analytical readouts and environmental conditions, in a structured, earchable environment. This creates a digital backbone that transforms tacit laboratory experience into regulator-ready knowledge and supports model development and validation, retrieval-augmented analytics and transparent decision support for chemistry, manufacturing and controls (CMC) teams. Together, these elements define an integrated vision for AI driven and robotics supported drug formulation. Using data-driven, multi-objective optimization, we consider performance and manufacturability in the same loop, guided by a patient-centred target product profile that treats ease of administration and tolerability as explicit design objectives alongside stability, dissolution behaviour and release profile. I will illustrate how this approach can fast-track preclinical research, improve efficiency across drug development pipelines and ultimately improve patient access to safe, effective and acceptable medicines.